scripts/to_parquet.py

"""Convert data/_train_raw.jsonl -> data/train-00000-of-00001.parquet
with a typed Arrow schema that HuggingFace `datasets` can load directly.
"""
from __future__ import annotations

import json
import sys
from pathlib import Path

import pyarrow as pa
import pyarrow.parquet as pq

REPO = Path(__file__).resolve().parent.parent
SRC = REPO / "data" / "_train_raw.jsonl"
DST = REPO / "data" / "train-00000-of-00001.parquet"

SEGMENT_TYPE = pa.struct([
    pa.field("speaker_id", pa.int32()),
    pa.field("timestamp", pa.string()),
    pa.field("text", pa.string()),
])

SCHEMA = pa.schema([
    pa.field("id", pa.string()),
    pa.field("source_collection", pa.string()),
    pa.field("source_file", pa.string()),
    pa.field("source_format", pa.string()),
    pa.field("topic", pa.string()),
    pa.field("round", pa.string()),
    pa.field("team_a", pa.string()),
    pa.field("team_b", pa.string()),
    pa.field("num_segments", pa.int32()),
    pa.field("num_chars", pa.int32()),
    pa.field("transcript", pa.string()),
    pa.field("segments", pa.list_(SEGMENT_TYPE)),
])


def main() -> int:
    rows = []
    with open(SRC, "r", encoding="utf-8") as f:
        for line in f:
            line = line.strip()
            if not line:
                continue
            rows.append(json.loads(line))

    if not rows:
        print("no rows in input", file=sys.stderr)
        return 1

    columns = {field.name: [] for field in SCHEMA}
    for r in rows:
        for field in SCHEMA:
            columns[field.name].append(r.get(field.name))

    table = pa.table(columns, schema=SCHEMA)
    pq.write_table(table, DST, compression="snappy")

    size = DST.stat().st_size
    print(f"wrote {table.num_rows} rows ({size:,} bytes) -> {DST}")
    print(f"schema:\n{table.schema}")
    return 0


if __name__ == "__main__":
    sys.exit(main())